Research map / Label-efficient and robust learning / Recognition and long-tail
Balanced Datasets Are Not Enough: Estimating and Mitigating Gender Bias in Deep Image Representations
Tianlu Wang, Jieyu Zhao, Mark Yatskar, Kai-Wei Chang, Vicente Ordonez
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add adversarial gender removal at conv5 | ResNet-50 · COCO | bias amplification (Delta): 9.93 -> 4.57 | Improved |
| add adversarial gender removal at conv5 | ResNet-50 · imSitu | bias amplification (Delta): 20.47 -> 6.63 | Improved |
| add adversarial gender removal at conv5 on balanced split | ResNet-50 (alpha=1) · COCO | bias amplification (Delta): 10.37 -> 2.51 | Improved |
| add adversarial gender removal at conv5 on balanced split | ResNet-50 (alpha=1) · imSitu | bias amplification (Delta): 21.63 -> 4.64 | Improved |
| add adversarial gender removal at conv5 | ResNet-50 · COCO | F1: 53.75 -> 52.54 | Hurt |
| add adversarial gender removal at conv5 | ResNet-50 · imSitu | F1: 40.11 -> 37.85 | Hurt |
| balance gender co-occurrence with labels (alpha=1) (ablation) | ResNet-50 · COCO | bias amplification (Delta): 9.93 -> 10.37 | No clear effect |
| balance gender co-occurrence with labels (alpha=1) (ablation) | ResNet-50 · imSitu | bias amplification (Delta): 20.47 -> 21.63 | No clear effect |
| move adversary from image to conv5 (ablation) | adversarial debiasing · COCO | bias amplification (Delta): 8.02 -> 4.57 | Improved |
| move adversary from conv4 to conv5 (ablation) | adversarial debiasing · COCO | bias amplification (Delta): 6.54 -> 4.57 | Improved |
| add adversarial debiasing to CRF model | CRF · COCO | bias amplification (Delta): 12.85 -> 4.81 | Improved |
| add RBA debiasing to CRF model | CRF · COCO | bias amplification (Delta): 12.85 -> 13.15 | No clear effect |
Closest papers by meaning
- Towards Fairness in Visual Recognition: Effective Strategies for Bias Mitigation (CVPR 2020 · similarity 0.91)
- Mitigating Gender Bias in Face Recognition using the von Mises-Fisher Mixture Model (ICML 2022 · similarity 0.91)
- Gender Artifacts in Visual Datasets (ICCV 2023 · similarity 0.89)
- Bias in Gender Bias Benchmarks: How Spurious Features Distort Evaluation (ICCV 2025 · similarity 0.89)
- Revealing and Reducing Gender Biases in Vision and Language Assistants (VLAs) (ICLR 2025 · similarity 0.85)
- Label-Efficient Group Robustness via Out-of-Distribution Concept Curation (CVPR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
